4.9 rating based on 350+ reviews

Written by Marijn Overvest | Reviewed by Sjoerd Goedhart | Fact Checked by Ruud Emonds | Our editorial policy

Gemini Gems for Procurement: 7 Custom Assistants to Build First

As taught in the Artificial Intelligence in Procurement course ★★★★★ 4.9 rating

Key takeaways

  • Gems are Gemini's equivalent of ChatGPT Custom GPTs and Claude Skills, named AI assistants with pre-configured scope and knowledge.
  • The seven procurement Gems worth building first match the seven ChatGPT Custom GPTs: supplier risk, contract review, negotiation prep, spend classifier, RFP scorer, category strategy, onboarding.
  • The design principles are the same across the three tools: tight scope, structured output, curated knowledge base, clear governance.

What Gemini Gems Are (and How They Differ from Custom GPTs)

Gemini Gems are named, reusable, capability-specific Gemini assistants. Where Gemini chat is the general-purpose assistant, a Gem is a specialised version configured for a specific workflow: a Contract Redline Reviewer Gem, a Supplier Risk Briefer Gem, a Category Brief Gem. Build once, invoke by name, share with the team.

Functionally, Gems are conceptually similar to ChatGPT's Custom GPTs and Claude's Skills. The mechanics differ in details (where they're configured, how they're shared, what knowledge sources they can access) but the underlying value pattern is the same: encode the team's recurring procedures into invokable assistants that any team member can use to produce comparable output.

For procurement teams operating in Google Workspace, Gems are the natural way to institutionalise the team's AI capability. The procurement function's standards become executable artefacts; new joiners can produce work at the team's standard from week two.

Seven Procurement Gems Worth Building First

In rough order of return on the build investment.

  • Contract Redline Reviewer — playbook deviation analysis on supplier-draft contracts
  • Supplier Risk Briefer — structured risk profile on a named supplier
  • Category Brief Generator — monthly or quarterly category narrative
  • RFP Response Scorer — structured extraction and scoring against the team's framework
  • Negotiation Prep Generator — the nine-question prep plus counterparty analysis
  • Savings Opportunity Surfacer — pattern detection in the spend file
  • Supplier QBR Brief Generator — the structured pre-QBR brief

Most procurement teams build these sequentially over 2-3 quarters. Each Gem benefits from the lessons learned on the prior build.

The Gem Build Pattern: From Idea to Working Assistant

The mechanics. A Gem has three components: instructions, knowledge sources, and (where applicable) connector actions.

Step 1, write the instructions. Treat the instruction like a brief to a new joiner. The role, the task, the inputs expected, the rules to apply, the output format, the uncertainty handling. Specific instructions produce consistent output; vague instructions produce variable output.

Step 2, attach knowledge sources. The team's playbook, the supplier list, the spend data references, the prior similar work as examples. The Gem references these in every invocation; the procurement team's context is built in rather than re-pasted each time.

Step 3, test on real work. Run the Gem on three to five real tasks. The first version is 70% right. Refine the instructions based on what's wrong. By the third iteration, the Gem is reliable enough for broader team use.

Step 4, share with the team. Once the Gem is reliable, share it via the team's Workspace. Document the owner, the use case, and any tips for invoking effectively. The Gem now becomes a team asset rather than an individual's tool.

Step 5, maintain. Each Gem has a quarterly review. Standards change, templates evolve, edge cases emerge. The owner refines the Gem; the team benefits.

Worked Example: Building the Contract Redline Reviewer Gem

The team's senior contracts manager builds the team's first Gem. The goal: encode the contract-review discipline that previously lived in the contracts manager's head.

The instructions: 'You are a contracts analyst reviewing supplier-draft contracts for a procurement team. For each draft contract: identify deviations from our standard playbook (attached as a knowledge source), classify each deviation by risk level (high, medium, low), and propose alternative wording aligned with our standard. Output as a structured table.'

The knowledge sources: the team's contract playbook, the team's standard fall-back positions per clause type, and three prior reviewed contracts as worked examples.

The first three test runs: the Gem correctly identifies 85% of deviations on the first pass, misses some less-common clauses, and over-flags some standard provisions as deviations. The contracts manager refines the instructions: adds the less-common clause patterns, clarifies what counts as standard.

By iteration 4: the Gem produces reviews comparable to the contracts manager's first-pass review. The team's contract reviews now start from a structured table instead of from a blank page. Per-contract review time drops from 60-90 minutes to 15-25 minutes.

Cumulative effect: the team handles 30% more contract reviews per quarter with the same headcount, and review consistency rises across team members as the Gem encodes the senior contracts manager's discipline. The Contract Management Course covers the playbook foundations this Gem encodes.

Governance, Team Sharing, and Library Maintenance

Three governance practices.

Named ownership. Each Gem has a procurement professional who owns it. The owner is responsible for the Gem's quality and for updates as standards evolve.

Library catalogue. The team maintains a one-page directory of available Gems: name, purpose, owner, last-updated date. Without this, Gems proliferate and team members don't know what's available. With it, new joiners learn the team's AI capability in their first week.

Quarterly review. Each quarter, the procurement function reviews the Gem portfolio. Which Gems get used regularly? Which are dormant? Which need refresh? The discipline keeps the library useful rather than letting it become a dusty collection.

Limits, and Where Gems Fall Short

Three honest limits.

Gem quality depends on instruction quality. A Gem with vague instructions produces vague output. The investment is in the instruction-writing; technical setup is the easy part.

Gems work best for structured-output tasks. Free-form thinking, exploratory analysis, and conversational discovery are still better as direct Gemini chat. Gems shine when the output has a defined structure.

Gems are not autonomous. They are invoked by humans for specific tasks. Procurement teams expecting Gems to run background work continuously will be disappointed; that's the Gemini equivalent of agentic workflows, which are a different capability.

Integrating Gems with the Rest of Google Workspace

The Gems library compounds with the Google Workspace integration.

Gems in Docs. Invoking a Gem within a Google Doc gives the Gem the Doc's context. The Contract Redline Reviewer Gem, invoked while viewing a draft contract in Docs, can read the contract directly.

Gems in Sheets. Invoking a Gem in Sheets gives the Gem the Sheet's data context. The Savings Opportunity Surfacer Gem, invoked while viewing the spend file, can analyse the data directly.

Gems with Drive knowledge sources. Knowledge sources in Drive (playbooks, templates, prior work) are persistently available to the Gem across invocations. The Gem references the current version of the playbook automatically.

Common Mistakes that Make Gem Libraries Underperform

Building Gems before standardising the procedure

A Gem encodes one person's view as if it were the team's standard. If the team has not agreed on the standard, the Gem effectively makes the call. Either standardise first, or build the Gem knowing the standardisation conversation is part of the build.

Skipping the worked-example layer

Gems with abstract instructions and no examples produce abstract output. Gems with two or three worked examples produce output that the senior team will accept. The examples matter more than the instruction prose.

Letting Gems become personal tools

A Gem in one person's account is not a team asset. Share it from day one. Procurement teams that treat Gems as personal productivity see no compounding value; teams that treat them as shared institutional capability see week-on-week improvement.

Not training new joiners on the library

The Gems library is institutional knowledge. New procurement professionals should learn the library in their first two weeks, not discover it six months in. Document a one-hour walkthrough; make it part of onboarding.

Want the templates and prompts from this article?

Every framework, template, and prompt referenced in this guide is included in our Artificial Intelligence in Procurement Course, ready to download and adapt for your team.

Frequently asked questions

How is a Gem different from a Custom GPT?

Similar in concept, different in implementation and ecosystem. Gems live inside Gemini; Custom GPTs live inside ChatGPT. Both provide scoped, reusable AI capability.

How long does it take to build a good Gem?

Four to six hours for a well-scoped Gem with a curated knowledge base. Longer for complex Gems with extensive knowledge.

How often should Gem libraries be reviewed?

Quarterly. Check the knowledge base is current, test output against standard scenarios, retire or rebuild as needed.

Ready to build this capability across your procurement team?

The AI Fundamentals for Procurement Teams program covers the prompt design, workflow structuring, and policy work that turn one-off wins into a durable AI capability.

Explore the program →